ShuriZma
/
Cxbx-Reloaded
mirror of https://github.com/Cxbx-Reloaded/Cxbx-Reloaded.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Merge pull request #1564 from NZJenkins/fix_warnings 

Fix warnings and broken WalkIndexBuffer bootstrap
This commit is contained in:
PatrickvL 2019-03-10 12:45:37 +01:00 committed by GitHub
parent d98f47c64c 0dce158641
commit 29e00d8f34
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
3 changed files with 28 additions and 51 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
src/core/hle/D3D8/Direct3D9/Direct3D9.cpp
View File

@ -7053,7 +7053,7 @@ void XTL::CxbxDrawIndexed(CxbxDrawContext &DrawContext)
//Walk through index buffer
// Determine highest and lowest index in use :
INDEX16 LowIndex, HighIndex;
WalkIndexBuffer_SIMD(LowIndex, HighIndex, &(DrawContext.pIndexData[DrawContext.dwStartVertex]), DrawContext.dwVertexCount);
WalkIndexBuffer(LowIndex, HighIndex, &(DrawContext.pIndexData[DrawContext.dwStartVertex]), DrawContext.dwVertexCount);
VertexBufferConverter.Apply(&DrawContext, LowIndex);
if (DrawContext.XboxPrimitiveType == X_D3DPT_QUADLIST) {
@ -7572,7 +7572,7 @@ VOID WINAPI XTL::EMUPATCH(D3DDevice_DrawIndexedVerticesUP)
else {
// Walk through the index buffer
INDEX16 LowIndex, HighIndex;
WalkIndexBuffer_SIMD(LowIndex, HighIndex, (INDEX16*)pIndexData, DrawContext.dwVertexCount);
WalkIndexBuffer(LowIndex, HighIndex, (INDEX16*)pIndexData, DrawContext.dwVertexCount);
// LOG_TEST_CASE("DrawIndexedPrimitiveUP"); // Test-case : Burnout, Namco Museum 50th Anniversary
HRESULT hRet = g_pD3DDevice->DrawIndexedPrimitiveUP(

55
src/core/hle/D3D8/Direct3D9/WalkIndexBuffer.cpp
View File

@ -5,33 +5,13 @@
#include <smmintrin.h> // SSE4.1
//#include <nmmintrin.h> // SSE4.2
//#include <immintrin.h> // AVX
#include "core\kernel\support\Emu.h"
#include "core\kernel\support\EmuXTL.h"
#include "common\util\CPUID.h"
#include "WalkIndexBuffer.h"
void WalkIndexBuffer_SSE41(XTL::INDEX16 & LowIndex, XTL::INDEX16 & HighIndex, XTL::INDEX16 * pIndexData, DWORD dwIndexCount);
// Walk an index buffer to find the minimum and maximum indices
void Init_SIMD
(
XTL::INDEX16 &LowIndex,
XTL::INDEX16 &HighIndex,
XTL::INDEX16 *pIndexData,
DWORD dwIndexCount
)
{
SimdCaps supports;
if (supports.SSE41())
WalkIndexBuffer_SIMD = WalkIndexBuffer_SSE41;
else
WalkIndexBuffer_SIMD = WalkIndexBuffer;
WalkIndexBuffer_SIMD(LowIndex, HighIndex, pIndexData, dwIndexCount);
}
void WalkIndexBuffer(XTL::INDEX16 & LowIndex, XTL::INDEX16 & HighIndex, XTL::INDEX16 * pIndexData, DWORD dwIndexCount)
// Default implementation
void WalkIndexBuffer_NoSIMD(XTL::INDEX16 & LowIndex, XTL::INDEX16 & HighIndex, XTL::INDEX16 * pIndexData, DWORD dwIndexCount)
{
// Determine highest and lowest index in use
LowIndex = pIndexData[0];
@ -45,6 +25,7 @@ void WalkIndexBuffer(XTL::INDEX16 & LowIndex, XTL::INDEX16 & HighIndex, XTL::IND
}
}
//SSE 4.1 implementation
void WalkIndexBuffer_SSE41(XTL::INDEX16 & LowIndex, XTL::INDEX16 & HighIndex, XTL::INDEX16 * pIndexData, DWORD dwIndexCount)
{
// We can fit 8 ushorts into 128 bit SIMD registers
@ -53,12 +34,13 @@ void WalkIndexBuffer_SSE41(XTL::INDEX16 & LowIndex, XTL::INDEX16 & HighIndex, XT
// Fallback to basic function if we can't even min / max 2 registers together
if (iterations < 2) {
WalkIndexBuffer(LowIndex, HighIndex, pIndexData, dwIndexCount);
WalkIndexBuffer_NoSIMD(LowIndex, HighIndex, pIndexData, dwIndexCount);
return;
}
__m128i *unalignedIndices = (__m128i*) pIndexData;\
__m128i min = _mm_set1_epi16(USHRT_MAX);
// Initialize mins and maxes
__m128i *unalignedIndices = (__m128i*) pIndexData;
__m128i min = _mm_set1_epi16(static_cast<short>(USHRT_MAX)); // cast as set1 only takes signed shorts
__m128i max = _mm_setzero_si128();
// Min / max over index data
@ -71,13 +53,13 @@ void WalkIndexBuffer_SSE41(XTL::INDEX16 & LowIndex, XTL::INDEX16 & HighIndex, XT
// horizontal min
min = _mm_minpos_epu16(min);
// horizontal max (using minpos)
max = _mm_subs_epu16(_mm_set1_epi16(USHRT_MAX), max); //invert
// horizontal max (no maxpos, we invert and use minpos)
max = _mm_subs_epu16(_mm_set1_epi16(static_cast<short>(USHRT_MAX)), max); //invert
max = _mm_minpos_epu16(max);
// Get the min and max out
LowIndex = (XTL::INDEX16) _mm_cvtsi128_si32(min);
HighIndex = (XTL::INDEX16) USHRT_MAX - _mm_cvtsi128_si32(max);
HighIndex = (XTL::INDEX16) USHRT_MAX - _mm_cvtsi128_si32(max); // invert back
// Compare with the remaining values that didn't fit neatly into the SIMD registers
for (DWORD i = dwIndexCount - remainder; i < dwIndexCount; i++) {
@ -89,4 +71,17 @@ void WalkIndexBuffer_SSE41(XTL::INDEX16 & LowIndex, XTL::INDEX16 & HighIndex, XT
}
}
// TODO AVX2, AVX512
// TODO AVX2, AVX512 implementations
// Detect SSE support to select real implementation on first call
void(*WalkIndexBuffer)(XTL::INDEX16 &, XTL::INDEX16 &, XTL::INDEX16 *, DWORD) =
[](XTL::INDEX16 &LowIndex, XTL::INDEX16 &HighIndex, XTL::INDEX16 *pIndexData, DWORD dwIndexCount)
{
SimdCaps supports;
if (supports.SSE41())
WalkIndexBuffer = WalkIndexBuffer_SSE41;
else
WalkIndexBuffer = WalkIndexBuffer_NoSIMD;
WalkIndexBuffer(LowIndex, HighIndex, pIndexData, dwIndexCount);
};

20
src/core/hle/D3D8/Direct3D9/WalkIndexBuffer.h
View File

@ -4,7 +4,7 @@
#include "core\kernel\support\Emu.h"
#include "core\kernel\support\EmuXTL.h"
void WalkIndexBuffer
extern void(*WalkIndexBuffer)
(
XTL::INDEX16 &LowIndex,
XTL::INDEX16 &HighIndex,
@ -12,22 +12,4 @@ void WalkIndexBuffer
DWORD dwIndexCount
);
void Init_SIMD
(
XTL::INDEX16 &LowIndex,
XTL::INDEX16 &HighIndex,
XTL::INDEX16 *pIndexData,
DWORD dwIndexCount
);
static void(*WalkIndexBuffer_SIMD)
(
XTL::INDEX16 &LowIndex,
XTL::INDEX16 &HighIndex,
XTL::INDEX16 *pIndexData,
DWORD dwIndexCount
) = Init_SIMD;
#endif